Oecologia

, Volume 168, Issue 3, pp 877–888

Stand- and tree-level determinants of the drought response of Scots pine radial growth

Authors

    • CREAF/Unitat d’Ecologia, Edifici C (Facultat de Ciències)Universitat Autònoma de Barcelona
    • School of GeosciencesUniversity of Edinburgh
  • Bernat C. López
    • CREAF/Unitat d’Ecologia, Edifici C (Facultat de Ciències)Universitat Autònoma de Barcelona
  • Lasse Loepfe
    • CREAF/Unitat d’Ecologia, Edifici C (Facultat de Ciències)Universitat Autònoma de Barcelona
  • Francisco Lloret
    • CREAF/Unitat d’Ecologia, Edifici C (Facultat de Ciències)Universitat Autònoma de Barcelona
Global change ecology - Original Paper

DOI: 10.1007/s00442-011-2132-8

Cite this article as:
Martínez-Vilalta, J., López, B.C., Loepfe, L. et al. Oecologia (2012) 168: 877. doi:10.1007/s00442-011-2132-8

Abstract

Characterizing the responses of key tree species to extreme climatic events may provide important information for predicting future forest responses to increased climatic variability. Here we aimed at determining which tree- and stand-level attributes were more closely associated with the effect of a severe drought on the radial growth of Scots pine, both in terms of immediate impact and recovery after the drought event. Our dataset included tree-ring series from 393 plots located close to the dry limit of the species range. Time series analysis and mixed-effects models were used to study the growth of each tree and its detailed response to a severe drought event that occurred in 1986. Our results showed that the radial growth responses of Scots pine were determined primarily by tree-level characteristics, such as age and previous growth rate, and secondarily by stand basal area and species richness, whereas local climate had a relatively minor effect. Fast-growing trees were more severely affected by the drought and retained proportionally lower growth rates up to three years after the episode. In absolute terms, however, fast-growing trees performed better both during and after the event. Older trees were found to be less resilient to drought. The effect of stand basal area and species richness indicated that competition for resources worsened the effects of drought, and suggested that the effect of interspecific competition may be particularly detrimental during the drought year.

Keywords

Extreme eventsPinus sylvestrisResilienceResistanceTree rings

Introduction

Climate change models predict an increase in the frequency and intensity of extreme climatic episodes (IPCC 2007), which could push individuals and communities beyond their stability thresholds and result in large-scale shifts in ecosystems (Scheffer et al. 2001). In the case of forests, particularly those in water-limited environments, an increase in the frequency and intensity of droughts is expected to become a key climatic stressor and the main agent of ecosystem change (Hartmann 2011). Recent reports of worldwide episodes of drought- and heat-related tree mortality (Allen et al. 2010) support this hypothesis and suggest that increased aridity is already affecting forests in many regions of the world. In this context, characterizing the responses of key species to current extreme events (Gaines and Denny 1993; Denny et al. 2009) may provide valuable information for predicting future community-level responses to increased climatic variability.

When characterizing the response of trees to extreme environmental conditions, it is important to distinguish between the immediate impact of the event and the recovery after its occurrence. This delayed response is frequently studied in terms of resilience, defined as the capacity of a system to recover its original state after disturbance (Holling 1973). As we use it here, the concept of resilience encompasses two different processes: resistance (related to the magnitude of the impact) and recovery (the speed of return to the original state) (cf. Côté and Darling 2010). Tree characteristics that confer resistance to a given disturbance (e.g., a drought event) are not necessarily the same as those that facilitate recovery after disturbance, and vice versa (e.g., Martín-Benito et al. 2008). It has been proposed, for instance, that slow growth may be associated with higher resistance to disturbance but, at the same time, it may compromise recovery, resulting in a trade-off between the resistance and recovery rates. Some support for this hypothesis is provided by studies of tree responses to browsing (Herms and Mattson 1992; Bee et al. 2007) and studies of cyclone damage to tropical rainforests (Curran et al. 2008).

The impact of a drought event on tree growth is determined by several factors, including the response of stomata, which determines the reduction in assimilation through stomatal closure during the drought, and the amount and use of stored carbohydrates, which could buffer the effect on growth of a given reduction in assimilation (e.g., McDowell et al. 2010). One could hypothesize, for instance, that trees characterized by high growth and maximum assimilation and transpiration rates under favorable conditions may have a stricter stomatal control when conditions become exceptionally dry (e.g., Oren et al. 1999). As a result, fast-growing trees would have more variable growth and lower resistance to extreme drought episodes, but a higher capacity for recovery as conditions return to normal. Alternatively, one could simply argue that high growth rates are more related to inherently superior genotypes or more suitable microenvironmental conditions. In this case, no trade-off would be expected, and fast-growing trees would be more resistant and would recover faster.

Other factors are likely to complicate the relationships described in the previous paragraph. Several studies have related tree age to differential growth responses, showing that in most cases old trees tend to be more sensitive to climate than younger age classes (e.g., Szeicz and MacDonald 1994; Carrer and Urbinati 2004; Rossi et al. 2008). If we confine ourselves to the impact of extreme droughts, both positive (Martínez-Vilalta and Piñol 2002; Lloret et al. 2004; Galiano et al. 2010) and negative (Mueller et al. 2005) effects of tree age on survival have been reported. The mechanism underlying these effects is unclear, as age per se (i.e., cellular senescence) does not appear to limit tree growth (Mencuccini et al. 2005) and may be mediated by size-related changes in the hydraulic system or the carbon balance of trees as they age (Ryan et al. 2006).

Stand-level properties also affect the drought responses of individual trees. High tree density and basal area have frequently been associated with increased drought-induced mortality (e.g., Bravo-Oviedo et al. 2005; Das et al. 2008; Galiano et al. 2010; but see Floyd et al. 2009 for a counterexample) and growth reductions (e.g., Klos et al. 2009; Linares et al. 2009; Vilà-Cabrera et al. 2011), indicating that competition may exacerbate the effects of water scarcity. Other community attributes, such as species richness, have been associated with stand-level stability in the face of disturbance (DeClerck et al. 2006; Lloret et al. 2007; Klos et al. 2009), the rationale being that species’ differential responses to environmental changes may lead to compensation and thus less variable aggregate properties in species-rich communities. At the individual tree level, species diversity may impact growth rates either positively by means of facilitation (e.g., hydraulic lift, Caldwell et al. 1998) or negatively through increased competition during times of limited resource supply (e.g., soil water preemption by deep-rooters, Casper and Jackson 1997). However, few studies have explored how species diversity affects stability in the face of disturbance at the individual tree level (Scherer-Lorenzen et al. 2007; Potvin and Gotelli 2008).

In this article, we study the response to drought of the radial growth of Scots pine (Pinus sylvestris L.) on 393 plots across the whole region of Catalonia (NE Spain), concentrating on the effects of a single extreme drought event. Radial growth provides a convenient way to characterize trees’ responses to disturbance, and it can be seen as an early warning of more severe effects, such as mortality (cf. Dobbertin 2005). Our main objective was to identify which tree attributes and plot-level properties are more closely associated with the impact of drought on tree growth, in terms of both resistance and recovery rate, during the years immediately following the event. More specifically, we aimed at determining (1) whether trees characterized by higher or more variable growth rates were more or less severely affected by drought; (2) whether older trees showed a lower capacity to confront and recover from drought episodes; and (3) to what extent climate and plot-level properties, such as basal area or species richness, modify individual trees’ responses to drought.

Materials and methods

The IEFC dataset

The collection of tree rings used in this study was obtained between 1988 and 1998 as part of the Catalan Ecological and Forest Inventory (IEFC, Burriel et al. 2000–2004; http://www.creaf.uab.es/iefc/). The study area is situated close to the southern (and dry) limit of the distribution of Scots pine and includes areas affected by drought-induced dieback (Martínez-Vilalta and Piñol 2002; Galiano et al. 2010). The IEFC sampled a total of 10,664 circular plots (radius = 10 m) distributed randomly throughout the forested area of Catalonia, NE Spain (1,214,408 ha of forest). The minimum distance between plots was 200 m. Scots pine was present on 3,219 plots (30.2%), and was the dominant tree species in terms of basal area on 1,962 plots (18.4%). Scots pine forests in Catalonia are mainly distributed on north-facing slopes between 800 and 1,600 m a.s.l. and cover an area of 219,754 ha. It is the second most abundant tree species in the region, after Pinus halepensis. On plots where Scots pine was dominant in terms of basal area, its average density was 903 trees ha−1, the average basal area was 21.6 m2 ha−1, the average canopy height was 13.1 m, and the average tree age was 49 years (Burriel et al. 2000–2004).

Tree rings were sampled in a random subset of approximately 20% of the IEFC plots. The dominant species was sampled on each of these plots (see Burriel et al. 2000–2004 for details). Cores were extracted with an increment borer from trees covering the whole tree diameter range represented on the plot. Trees within a plot were selected to represent a given diameter class (5 cm range), and these were sampled in proportion to their abundance on the plot. In each tree, one single core extending along the whole stem was extracted from N to S at a height of approximately 0.5 m. The cores were placed on wooden supports and taken to the laboratory for analysis. In the laboratory, all the cores were air-dried, fixed to the support and smoothed by sanding with progressively finer grade sandpaper until the growth rings could be easily recognized. Ring width was measured to the nearest 0.01 mm using a binocular scope and a linear table attached to a PC (CATRAS system; Aniol 1983).

Data selection and quality control

From the original IEFC dataset, we selected all plots on which Scots pine was present and cores had been sampled from at least four Scots pine trees. The software COFECHA (Grissino-Mayer 2001) was used to cross-date the samples and check the consistency of the different ring-width time series within plots. Only those trees showing positive correlations with the other trees on the plot (r > 0.1) were kept for further analysis, and the dataset was filtered again so that only plots with at least three consistent growth series were retained. A final filtering was carried out to remove the trees that were very small at the time of sampling (diameter at breast height, d.b.h. <10 cm) and those for which the reconstruction of past diameters led to inconsistent results (e.g., because the wood cores did not reach the center of the tree). The final database contained 1,433 trees from 393 plots, spanning the period 1751–1997.

Climatic data

A climatic dataset was created for each sampled plot by modeling climatic surfaces from discrete climatic data from the Spanish weather monitoring system (National Weather Institute; http://www.inm.es). All weather stations in Catalonia with available data were included in the analysis. The time period was limited to January 1951 to December 1998, since before 1951 there were insufficient weather stations to achieve reasonable accuracy in model predictions. Multiple regressions with residual correction were used, together with spatial interpolation based on inverse distance weighting and splines, to predict monthly and annual values of precipitation and minimum, maximum, and mean temperatures with a spatial resolution of 180 m (cf. Ninyerola et al. 2007a, b for a detailed account of the methodology).

Monthly potential evapotranspiration (PET) was calculated from monthly values of minimum, maximum, and mean temperatures using the Hargreaves–Samani method (1982), which estimates solar radiation from temperature amplitude. The average values (1951–1998) of mean annual temperature (T), annual precipitation (P), annual PET, and the average ratio of precipitation to PET for June–August (P/PETSUM, a measure of summer drought stress) were used to characterize the climate of each plot. All annual values were calculated for the period from August of the year before the formation of the tree ring to July of the year of ring formation, in accordance with the growth–climate response functions of the study species (Gutiérrez 1989). The studied plots covered a wide range of climatic conditions, with average annual temperatures ranging from 5.5 to 13.9°C, annual precipitation ranging between 580 and 1,280 mm, and summer P/PET between 0.29 and 1.04.

We studied tree responses to one individual drought event that occurred in 1986. This drought represented, along with those occurring in 1991 and 1994, one of the driest episodes in the period 1951–1998 in terms of summer P/PET (Fig. 1), and it affected the whole study region (1986 was among the three years with lowest summer P/PET on 91% of the study plots). The 1985–1986 drought was associated with widespread detrimental effects on the vegetation in many regions of Spain (Camarero et al. 2004; Lloret et al. 2004). The main advantage of concentrating on one common drought event, instead of studying the responses to the driest event specific to each plot, is that it makes the environmental conditions (e.g., climate) before and after the event much more comparable across plots, thus facilitating the comparison of growth responses. The reason why we chose the 1986 drought, instead of the even more extreme episode in 1994, is that most of the plots were sampled before or during 1994. Therefore, studying the 1994 event would have greatly limited our sample size.
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-011-2132-8/MediaObjects/442_2011_2132_Fig1_HTML.gif
Fig. 1

Temporal dynamics of temperature, precipitation, summer P/PET (ratio of precipitation to potential evapotranspiration), and basal area increment (BAI) at the study plots for the period 1952–1996. Year 1986, corresponding to the studied drought episode, is indicated by an asterisk in each panel. Boxes indicate the lower and upper quartiles, the band inside the box is the median, and whiskers extend to the most extreme data points that are within 1.5 times the interquartile range

Growth modeling and data analyses

Basal area increment (BAI) was used to characterize tree growth. BAI was calculated from ring-growth according to
$$ {\text{BAI}}_{t} = \pi \cdot \left( {R_{t}^{2} - R_{t - 1}^{2} } \right), $$
(1)
where R is the radius of the tree and t the year of tree ring formation. We characterized tree growth and its general drought sensitivity by modeling individual BAI as a function of yearly climate, tree- and plot-level attributes using a linear mixed-effects model:
$$ \begin{aligned} \ln (BAI_{{i\left( {jk} \right)}} ) =& \beta_{0} + \gamma_{0i} + \alpha_{0i\left( j \right)} + \beta_{1} \cdot \ln \left( {{\text{age}}_{{i\left( {jk} \right)}} } \right) \\ & + \beta_{2} \cdot \ln \left( {{\text{d}} . {\text{b}} . {\text{h}} ._{{i\left( {jk} \right)}} } \right) + \beta_{3} \cdot {\text{year}}_{k} + \beta_{4} \cdot {\text{temperature}}_{ik} \\ & + \beta_{5} \cdot \ln \left( {P / {\text{PET}}_{{{\text{SUM}}ik}}} \right) + \beta_{6} \cdot {\text{temperature}}_{ik} \cdot \ln \left( {P / {\text{PET}}_{{{\text{SUM}}ik}}}\right) \\ & + \beta_{7} \cdot \ln \left( {{\text{age}}_{i(jk)} } \right) \cdot \ln \left( {P / {\text{PET}}_{{{\text{SUM}}ik}}} \right) \\ & + \beta_{8} \cdot {\text{management}}_{i}+ \beta_{ 9} \cdot {\text{richness}}_{i} + \beta_{ 1 0} \cdot {\text{sqrt}}\left( {{\text{BA}}_{i} } \right) \\ & + \beta_{ 1 1} \cdot \ln \left( {P / {\text{PET}}_{{{\text{SUM}}ik}} } \right) \cdot {\text{sqrt}}\left( {{\text{BA}}_{i} } \right) +\varepsilon_{{i\left( {jk} \right),}}\\ \end{aligned} $$
(2)
where β0 is the overall intercept and β1 to β11 are parameters adjusting the fixed effects (see Table 1 for a description of all the fixed factors included in the model), i is the index for plot, i(j) is the index for tree nested in plot, k is the index for the year of measurement, γ0 and α0 are the random effects on the intercept associated with plot and tree, respectively, and εi(jk) is the error term. Temporal autocorrelation of the error term was accounted for using an ARMA(1,1) correlation structure. Random effects were assumed to follow a normal distribution with zero mean. The response variable was normalized by applying natural logarithms because the distribution of BAI was strongly skewed to the left. The variable year is best seen as a composite of many directional changes occurring during the second half of the twentieth century, such as increasing CO2 concentration in the atmosphere and N deposition. Including the variable year resulted in significantly better model fits than those obtained from alternative and highly correlated variables such as yearly [CO2].
Table 1

Variables included as fixed factors in the mixed-effects models of BAI (growth) and its immediate (impact) and delayed (delay) responses to the 1986 drought (cf. Eqs. 2, 3)

Variable

Description

Units

Range

Model

Tree-level factors

 d.b.h.

Current diameter at breast height

cm

10–55.9

Growth

 Age

Current tree age

year

2–175

Growth

 Age1986

Tree age in 1986

year

8–166

Impact, delay

 Impact

Magnitude of growth reduction in 1986

−3.7 to 2.8

Delay

 BAIPRE

Average BAI for the period 1981–1985

cm2

0.3–55.2

Impact, delay

Climatic and temporal factors

 Temperature

Mean annual temperature from August of previous year to July of current year

°C

3.7–15.2

Growth

 P/PETSUM

Yearly ratio of June–August precipitation to potential evapotranspiration

0.04–2.36

Growth

 P/PETSUM,P

Plot average of P/PETSUM (1952–1996)

0.29–1.04

Impact, delay

 P/PETREL

Ratio of yearly P/PETSUM to P/PETSUM,P for years 1986, 1987, 1988, or 1989

1986: 0.3–0.6

Impact, delay

1987: 0.6–1.6

1988: 0.4–1.1

1989: 0.4–1.3

 Year

Current year

1952–1996

Growth

Plot-level factors

 Management

Presence of forest management

Yes/no

Growth, impact, delay

 BA

Plot basal area

m2 ha−1

0.9–87.4

Growth, impact, delay

 Richness

Tree species richness (plot-level)

Discrete (3 levels)

Growth, impact, delay

Tree species richness ranged between 1 (i.e., monospecific Scots pine stands) and 8, and showed a very uneven distribution. Consequently, it was introduced as a discrete variable with three levels in all models [(1) 1 species, N = 103 plots; (2) 2–3 species, N = 136 plots; and (3) 4–8 species, N = 149 plots]. Forest management could influence both the overall growth patterns of the studied trees and their response to a particular drought event. In order to account for these potential effects, we introduced a dichotomic variable in our models, based on the field observations recorded during the IEFC sampling. A forest was deemed “managed” if at the time of sampling there was any evidence of current or past forest management or of major land-use changes (including tree cutting and the presence of terraces indicating land abandonment). Approximately two-thirds of the study plots (64%) were considered managed according to this criterion. Identical results were obtained in all models if management was defined in a more restrictive manner to include only plots where evidence of tree cutting was observed during the IEFC (not shown).

To assess the effect of a specific drought event, series of BAI for each individual tree were modeled as time series using ARIMA models (Cook 1987). The square root transformed BAI was used, as this transformation improved the normality of the residuals of the fitted models. The best ARIMA(p,d,q) model (in terms of the Akaike information criterion, AIC) was fitted to the BAI time series of each individual tree, accounting for innovative outliers (Cryer and Chan 2010). The maximum order of the fitted ARIMA models was limited to p = 1, d = 1, q = 1, as higher order models are much more difficult to interpret and are rare in tree ring time series (cf. Downing and McLaughlin 1992; Druckenbrod 2005). ARIMA models were successfully fitted to 1,382 out of 1,433 individual tree time series. On average, the individual ARIMA models accounted for 60% of the variation in the raw BAI time series. Since we were interested in studying the responses of trees to one individual drought event that occurred in 1986, ARIMA models for each individual tree were only fitted up to year 1985. For the following years, the difference between measured BAI values and those predicted by the fitted ARIMA models were used as measures of how much growth for a given year departed from the expected value given the overall BAI time series.

The standardized 1986 difference between the measured BAI and that predicted by the ARIMA model fitted to each individual time series was used as a measure of the immediate impact of the 1986 drought on growth (referred to as “impact” hereafter). This variable was highly correlated to more conventional measures of drought impact (e.g., r = 0.75 with the growth during the drought event divided by the average growth during the preceding 5 years), but was more satisfactory in terms of its distribution (normality) and the fit of the resulting models. Similarly, the delayed impact of the drought was measured as the standardized difference between measured and predicted BAI for the years immediately after the drought event (Delayed_1, Delayed_2, and Delayed_3 corresponded to years 1987, 1988, and 1989, respectively) (cf. Orwin and Wardle 2004). The use of standardized values (difference divided by the square root of the time series variance) as a response variable has the advantage of accounting for inherent differences in growth variability across trees, and thus makes it possible to separate the effects of growth rate from those of growth variability.

In order to characterize the factors determining the different responses of individual trees to the 1986 drought, we used linear mixed-effects models with impact, Delayed_1, Delayed_2, and Delayed_3 modeled as a function of tree-level and plot-level explanatory variables:
$$ \begin{gathered} {\text{Response}}_{i(j)} = \beta_{0} + \gamma_{0i} + \beta_{1} \cdot \ln \left( {{\text{age}}_{{1986_{i(j)} }} } \right) + \beta_{2} \cdot \ln \left( {{\text{BAI}}_{{{\text{PRE}}_{i(j)} }} } \right) + \beta_{3} \cdot P / {\text{PET}}_{{{\text{SUM,P}}i}} + \beta_{4} \cdot P / {\text{PET}}_{{{\text{REL}}i}} + \hfill \\ \beta_{5} \cdot {\text{management}}_{i} + \beta_{ 6} \cdot {\text{richness}}_{i} + \beta_{ 7} \cdot {\text{sqrt}}\left( {{\text{BA}}_{i} } \right) +\varepsilon_{i(j)}, \hfill \\ \end{gathered} $$
(3)
where β0 is the overall intercept and β1 to β7 are parameters adjusting the fixed effects (see Table 1 for a description of all fixed factors included in the model), i is the index for plot and i(j) the index for tree, γ0 is the random effect associated with plot, and εij is the error term. The variable BAIPRE characterized growth immediately before the event (average BAI for the 1981–1985 period) and was highly correlated with tree d.b.h. in 1986. The variable impact was added as an additional fixed factor in the models describing the delayed effects of the 1986 drought. Plot climatic attributes were characterized using two variables: average summer P/PET (P/PETSUM,P) and the severity of the summer drought in the modeled year (1986, 1987, 1988, or 1989) in terms of summer P/PET (P/PETREL). Species richness and the presence of forest management were coded as explained before.
All analyses were conducted with the software R (v. 2.8, the R Foundation for Statistical Computing) using the packages TSA and forecast for time-series analyses and the package nlme for linear mixed-effects modeling. Quantitative variables were log or square root transformed when required to improve their normality or to linearize relationships (cf. Tables 2, 3, 4). A correlation matrix was used to select explanatory variables so that highly correlated pairs of variables were avoided. In all of the final models the variance inflation factor (VIF) of all variables in the fixed part of the model was <3. The residuals of the mixed-effects models described above showed no pattern. Significance for all statistical analyses was accepted at α = 0.05. The R2 (explained variance) of the mixed models was estimated using a likelihood ratio statistic (Magee 1990).
Table 2

Summary of the mixed-effects model of BAI (period 1952–1996)

Fixel effect

Estimate

SE

df

t value

P value

Intercept

5.403

1.471

46,316

3.673

<0.001

ln (d.b.h.)

1.894

0.017

46,316

108.896

<0.001

ln (Age)

−0.602

0.014

46,316

−43.713

<0.001

Temperature

0.006

0.003

46,316

1.985

0.047

ln (P/PETSUM)

0.012

0.033

46,316

0.357

0.721

Year

−0.003

0.001

46,316

−4.100

<0.001

Management

0.007

0.031

385

0.107

0.915

sqrt(BA)

−0.113

0.013

385

−8.800

<0.001

Richness = 2

−0.011

0.034

385

0.339

0.735

Richness = 3

0.009

0.034

385

0.251

0.802

ln(Age) × ln(P/PETSUM)

0.004

0.005

46,316

0.858

0.391

Temperature × ln(P/PETSUM)

0.023

0.002

46,316

10.343

<0.001

sqrt(BA) × ln(P/PETSUM)

−0.023

0.004

46,316

−6.384

<0.001

The random part of the model includes both plot- and tree-level (nested within plot) effects. Model R2 = 0.89

See Table 1 for a description of all variables and acronyms

Bold characters indicate significant effects (P < 0.05)

Table 3

Summary of the mixed-effects model of the impact of the 1986 drought on BAI

Fixel effect

Estimate

SE

df

t value

P value

Intercept

0.106

0.985

925

0.108

0.914

ln(Age1986)

−0.127

0.102

925

−1.250

0.212

ln(BAIPRE)

−0.171

0.049

925

3.468

<0.001

Management

−0.290

0.174

376

−1.662

0.097

sqrt(BA)

0.266

0.076

376

3.515

<0.001

Richness = 2

−0.064

0.217

376

−0.295

0.768

Richness = 3

0.509

0.222

376

2.287

0.023

P/PETSUM,P

1.742

0.749

376

2.327

0.021

P/PETREL

−0.839

1.443

376

−0.581

0.561

Model R2 = 0.25

Bold characters indicate significant effects (P < 0.05)

See Table 1 for a description of all variables and acronyms

Table 4

Summary of the mixed-effects model of the delayed effects of the 1986 drought on BAI

Fixed effect

1 year delay (1987)

2 year delay (1988)

3 year delay (1989)

Estimate

SE

P value

Estimate

SE

P value

Estimate

SE

P value

Intercept

1.968

0.492

<0.001

2.507

0.703

<0.001

1.494

0.781

0.056

Impact

0.165

0.023

<0.001

0.261

0.024

<0.001

0.263

0.027

<0.001

ln(Age1986)

−0.378

0.082

<0.001

−0.224

0.085

0.008

0.121

0.097

0.213

ln(BAIPRE)

−0.289

0.044

<0.001

−0.364

0.045

<0.001

−0.355

0.053

<0.001

Management

−0.085

0.118

0.474

0.066

0.120

0.581

0.074

0.137

0.592

sqrt(BA)

−0.176

0.052

<0.001

−0.128

0.054

0.018

−0.226

0.062

<0.001

Richness = 2

0.217

0.145

0.136

0.120

0.148

0.417

0.089

0.169

0.599

Richness = 3

0.012

0.150

0.935

0.053

0.154

0.730

0.080

0.175

0.648

P/PETSUM,P

0.170

0.454

0.709

−0.999

0.455

0.029

0.816

0.521

0.118

P/PETREL

0.325

0.339

0.338

0.249

0.810

0.759

0.177

0.638

0.782

R2 = 0.17, 0.24 and 0.17 for the models corresponding to 1987, 1988, and 1989, respectively. Sample sizes as in Table 3

Bold characters indicate significant effects (P < 0.05)

See Table 1 for a description of all variables and acronyms

Results

BAI as a function of yearly climate, tree attributes, and plot-level characteristics

The initial model relating BAI to yearly climate and plot- and tree-level attributes explained a large fraction of the variance of the growth data (R2 = 0.89). Most of the variance in BAI was accounted for by tree properties (Table 2): growth increased with tree size (d.b.h.) and declined with tree age. Including these two variables in the model gave similar growth trajectories and a better model fit than using a quadratic function of d.b.h., as we did in a previous paper (Martínez-Vilalta et al. 2008). Significant climatic variables included temperature and its interaction with P/PETSUM. This interaction implied that when summer P/PET was high (wet conditions), temperature had a slightly positive effect on BAI, whereas when summer P/PET was low (dry conditions), increasing temperature had the opposite effect (cf. Electronic supplementary material 1).

Tree basal area was the only plot-level variable with a significant effect on the model, with lower Scots pine BAI in plots with higher total basal area (Table 2). Basal area interacted with P/PETSUM, so that the increase in BAI with P/PETSUM was particularly noticeable on sites with relatively low BA, whereas growth was similarly low, regardless of P/PETSUM, on sites with high BA (cf. Electronic supplementary material 2). Including additional interaction terms with plot-level variables (Management × P/PETSUM, Richness × P/PETSUM) worsened the model fit in terms of AIC.

Impact of and recovery from 1986 drought

The 1986 residuals of the BAI time-series were negative for most trees (Fig. 2a), indicating that growth during this drought year was below average in most cases. The same pattern was apparent for the 1987 and 1988 residuals, although they were, on average, closer to zero than those corresponding to 1986 (Fig. 2b, c). By 1989, the residuals were still slightly negative on average, but they were centered approximately around zero (Fig. 2d), implying that growth was not lower than expected, so the trees had recovered by roughly three years after the drought.
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-011-2132-8/MediaObjects/442_2011_2132_Fig2_HTML.gif
Fig. 2

Histograms of the standardized differences between measured and predicted BAI, corresponding to the year of the drought (1986, a) and the three subsequent years (1987–1989, bd). These values were used as response variables in the models aimed at explaining the impact and recovery after the 1986 drought (see text)

The impact of the 1986 drought on growth (Table 3; model R2 = 0.25) was associated with the growth rate prior to the drought event (BAIPRE), with greater drought impacts on fast-growing trees. Across trees, BAIPRE was positively related to the variability of the time series residuals (Fig. 3), implying that faster growth was associated with higher growth variability. Basal area was the plot-level variable with the greatest effect on the model. Plots with higher basal area were less affected by the drought, whereas high species richness and high average summer P/PET at plot level increased the impact of the drought effect. Although there was no significant interaction between species richness and basal area, the effect of basal area changed sign (to significantly negative) if species richness was removed from the model (not shown). Finally, the presence of management had a marginally significant effect, also increasing drought impact.
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-011-2132-8/MediaObjects/442_2011_2132_Fig3_HTML.gif
Fig. 3

Relationship between average basal area increment (BAI) in the period 1981–1985 (BAIPRE) for a given tree and the variance of its residuals in the time series BAI model (see text). The regression line and corresponding R2 are also shown

The models of the delayed effects of the 1986 drought explained a slightly lower proportion of the variance in the dataset compared to the model describing the immediate impact of the drought (R2 = 0.17, 0.24, and 0.17 for the 1987, 1988, and 1989 models, respectively). Tree age and previous BAI had a negative effect on recovery one year after the drought (Table 4). The magnitude of the drought impact was also significant, in the sense that trees that had been more severely affected showed lower growth one year after the drought. The effect of tree age disappeared three years after the drought, whereas the effects of impact and previous BAI lasted until 1989 (Table 4). Our results imply that fast-growing trees took longer to recover, despite the fact that they retained higher absolute growth rates both during and after the drought (Fig. 4). The effect of BA was negative, implying that recovery was slower on plots with high basal area. The plot-level average of summer P/PET had a significant negative effect in 1988, coinciding with a relatively dry summer (Fig. 1). The effects of management, species richness and the relative climatic dryness (P/PETREL) were not significant in any of the models.
https://static-content.springer.com/image/art%3A10.1007%2Fs00442-011-2132-8/MediaObjects/442_2011_2132_Fig4_HTML.gif
Fig. 4

Relationship between average basal area increment (BAI) in the pre-drought period (1981–1985; BAIPRE) and a BAI during the year of the drought (1986 BAI) or b average BAI during the period immediately after the drought (post-drought BAI, period 1987–1989). The regression lines and corresponding R2 are also shown

Discussion

Fast-growing trees are more severely affected but recover faster

Our results agree with previous studies showing that dominant or fast-growing trees suffer greater growth reductions in the event of a disturbance (e.g., Martín-Benito et al. 2008), and with the notion that the recovery rate after disturbance is also higher in fast-growing trees (e.g., Curran et al. 2008). The fact that tree growth before, during, and after the drought episode were highly correlated indicates that faster-growing trees continue to perform better after a stressful period, and suggests that competitive relationships at the intraspecific level were not fundamentally altered as a result of the drought episode. However, our study shows that fast-growing trees retained proportionally lower growth rates up to three years after the drought, and their resilience was therefore lower than that of slow-growing trees. In other words, the positive correlation between growth rates before and after the drought (Fig. 4b) was not enough in our case to compensate for the disproportionately larger initial growth decline experienced by fast-growing trees.

The physiological mechanisms underlying the higher drought vulnerability of fast-growing trees deserve further attention, as the implications for vegetation dynamics can be substantial in the current context of increasing drought frequency and intensity. The differences observed are likely to reflect either growth-related differences in water or carbon use during and after the drought, involving a proportionally greater impact of drought on leaf area or carbon reserves (Galiano et al. 2011) in fast-growing trees, or their higher metabolic costs, which may represent a proportionally larger burden during periods of low water availability. Alternatively, and considering that faster-growing trees were also larger, at least in terms of d.b.h., the pattern observed may also be related to indirect effects of tree size. Tall trees experience more severe water deficits, due to their longer hydraulic path length and higher hydraulic resistance (Domec et al. 2008), and they may have a lower return in carbon gain from their investment in stem and leaf biomass (Zhang et al. 2009).

Older trees grow less and recover more slowly

Our results show that, for a given tree growth rate, older individuals recovered more slowly after the 1986 drought. This result is consistent with previous studies showing that old trees are more sensitive to climate under stressful conditions, particularly in high-altitude treeline environments (Szeicz and MacDonald 1994). This pattern has been attributed to the increased hydraulic costs of tree height (Carrer and Urbinati 2004) and the delayed basipetal movement of growth-initiating hormones in taller and older trees (Rossi et al. 2008). The situation may be more complex, however, under Mediterranean conditions, where water scarcity tends to be the main stress factor for trees. Rozas et al. (2009), for instance, concluded that climatic sensitivity decreased with increasing age in Juniperus thurifera growing in north-central Spain, a result they attribute to greater root development in older trees. Similarly, Vieira et al. (2009) found that earlywood and tree-ring width of younger Pinus pinaster were more sensitive to climate fluctuations, whereas the response of latewood to climate was stronger in older trees.

As suggested in the previous section, however, the effects of tree size, age, and growth rate are frequently mixed and, in practice, they are difficult to separate unequivocally in field studies of tree populations. We tried to account for this by including a measure of tree size (d.b.h.) and age in our growth model (Table 2), and previous growth rate and age in the impact and delayed-effects models (Tables 3, 4). The results of the growth model showed that, for a given tree size, older trees grew less. This is not surprising if one considers that the current d.b.h. of a tree is, in actual fact, its cumulative radial growth over its lifetime (i.e., over a time period equal to its age). In any case, it is likely that the effects we attribute to age here are not direct age effects (cf. Mencuccini et al. 2005), but indirect effects of age-related changes in tree size or allometry (e.g., stature, belowground allocation) not captured entirely by differences in d.b.h. or radial growth. Additionally, time-dependant factors, such as the increase in the cumulative probability of trees being damaged by abiotic factors (wind, snow) or affected by pathogens that act as carbon sinks, may also explain why older trees grew proportionally less and took longer to recover.

Effect of stand basal area and species richness

Basal area was the main stand-level factor affecting BAI and growth responses to the studied drought episode. Trees in stands with higher basal area grew less, and they also showed a more perdurable growth reduction in response to the 1986 drought, consistent with previous research (cf. Guarín and Taylor 2005; Klos et al. 2009; Linares et al. 2009). Moreover, species richness had no effect on overall growth or on recovery after drought, in agreement with previous studies in the same study area focusing on stand-level productivity (Vilà et al. 2003) and normalized difference vegetation index (NDVI) reductions after a drought event (Lloret et al. 2007). During the drought year, however, the situation appeared quite different: a significant effect of species richness indicated that pines on stands with more tree species suffered greater drought effects on growth (Table 3), while, unexpectedly, trees in stands with higher basal area tended to be less affected. In principle, the lower drought impact on trees in plots with higher basal area would be consistent with the prediction that positive interactions are more common under high abiotic stress (the “stress-gradient hypothesis,” Bertness and Callaway 1994), and with previous studies showing temporal changes in the net direction of plant–plant interactions (e.g., Kitzberger et al. 2000; Sthultz et al. 2007). We believe, however, that in this case the effect of the basal area is partially confounded by species richness. This interpretation is supported by the fact that the effect of basal area on drought impact became negative (greater impact on plots with higher basal area) if species richness was not included in the model.

It should be noted that the effects of species richness and basal area may be explained by a covariation with (unmeasured) soil conditions if, for instance, more diverse stands tend to occur on edaphically drier sites. At the same time, however, the composition of the studied communities may also play an important role. Since the study area is on the dry edge of the distribution of Scots pine, co-occurring tree species (mostly Quercus ilex, Q. humilis, Q. faginea, and Pinus nigra, according to the IEFC dataset; Burriel et al. 2000–2004) tend to have deeper root systems (Canadell et al. 1996) and are considered more drought resistant than Scots pine (Martínez-Vilalta and Piñol 2002; Marañón et al. 2004). Thus, the number of competing, drought-resistant species is likely to increase with the total number of species in the community, potentially explaining the negative effect of species richness on Scots pine growth during a drought year. Taken as a whole, our results suggest that the effects of interspecific competition were stronger than those of intraspecific competition during the drought event, whereas the effect of competition during the recovery period after drought was encapsulated by basal area, an aggregated measure that does not distinguish between the identities of species.

Humid locations may be highly vulnerable to drought events

The direct effect of climate on the growth response of Scots pine to the 1986 drought was relatively minor, suggesting that site-to-site differences in drought responses were more closely related to stand structure than to climatic differences across localities. This result contrasts with the importance of climate in determining absolute growth rates in Scots pine (Table 2, Martínez-Vilalta et al. 2008). It should be noted, however, that other factors, such as different seasonality of rainfall or soil characteristics across sites, are likely to influence water availability and drought responses at the local level (e.g., Lapointe-Garant et al. 2010). In addition, it has been shown that growth responses of trees may integrate long-term rainfall regimes, particularly during dry periods, reflecting increased dependence on deep soil water (Sarris et al. 2007). These factors were not included in our analyses, and they probably underlie the relatively low variance explained by our models of the effects of the 1986 drought (Tables 3, 4).

The fact that average summer P/PET had a significant negative effect in the dry years 1986 and 1988, whereas there was no effect during the wetter years 1987 and 1989 (Tables 3, 4; Fig. 1), suggests that tree growth in more humid localities may be inherently more vulnerable to drought. This is consistent with the idea that adjustments along aridity gradients at the tree and stand levels (Mencuccini 2003; Martínez-Vilalta et al. 2009; Vilà-Cabrera et al. 2011) may be highly effective in compensating for differences in water availability. In the case of Scots pine, these adjustments involve major changes in the leaf-to-sapwood area ratio (Martínez-Vilalta et al. 2009), and should be taken into account when predicting large-scale growth trends under climate change scenarios (Reich and Oleksyn 2008). It should be noted, however, that compensation frequently occurs at the expense of lower average growth rates in drier locations, as shown in this study, and that many other studies have clearly shown that extreme drought effects (i.e., mortality) tend to be more common on drier sites (Gitlin et al. 2006; Vilà-Cabrera et al. 2011).

In conclusion, our study shows that the radial growth responses of Scots pine to a specific drought event were determined primarily by tree-level characteristics, such as age and previous growth rate, and only secondarily by stand basal area and species richness. The direct effect of local climate, in contrast, was relatively minor. Trees with higher growth rates prior to the drought were proportionally more impacted by the episode and were still more affected three years after the event. In absolute terms, however, faster growing trees performed better both during and after the stressful event, indicating that competitive relationships at the population level were not fundamentally altered as a result of the drought. This pattern has important implications for the prediction of the impacts of climate change type droughts on forests. However, an important challenge remains to be confronted: that of translating tree responses in terms of growth reductions into true demographic impacts in terms of mortality (cf. Pedersen 1998; Bigler and Bugmann 2004), reproductive success, and species interactions.

Acknowledgments

We thank the IEFC staff for collecting and processing all the samples, Rosa Maria Roman Cuesta for early discussions on the ideas developed in this study, Jordi Vayreda for helping search the IEFC database, and Llorenç Badiella for providing statistical advice. We would also like to thank Henrik Hartmann, an anonymous reviewer, and the Editors for their useful suggestions on an earlier version of the manuscript. Miquel Ninyerola provided the gridded climatic datasets obtained in the framework of project CGL2006-01293, funded by the Spanish Ministry of Science and Innovation. This study was supported by the Spanish Ministry of Science and Innovation via competitive projects CGL2007-60120, CSD2008-0004 and CGL2009-0818.

Supplementary material

442_2011_2132_MOESM1_ESM.doc (188 kb)
Supplementary material 1 (DOC 188 kb)
442_2011_2132_MOESM2_ESM.doc (190 kb)
Supplementary material 2 (DOC 189 kb)

Copyright information

© Springer-Verlag 2011